Innovator-VL-RL-172K: 172,000 Multimodal Reinforcement Learning Instances

Innovator-VL-RL-172K is a curated multimodal reinforcement learning dataset containing 172,000 instances released by InnovatorLab in 2026. It provides image-text reasoning pairs designed to support RLHF-style optimization for vision-language models.

Use Cases

• Optimizing vision-language models using the 172,000 image-text reasoning pairs for RLHF
• Benchmarking decision-making performance in multimodal agents using curated response selection samples
• Fine-tuning models to improve visual reasoning accuracy through reinforcement learning optimization

Strengths

• 172,000 curated instances
• MIT licensed
• Linked to Arxiv research 2601.19325

Limitations

• Lack of detailed column descriptions in metadata
• Potential geographic bias toward US-centric data
• Curated nature may introduce selection bias

Provenance

Source

InnovatorLab

Collection Method

Curated

Freshness

Last updated February 2026.

Geography

United States